Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

A Simulation of Project Completion Probability

Using Different Probability Distribution Functions

Erimas Tesfaye, Kidist Girma, Eshetie Berhan, and Birhanu Beshah

Addis Ababa University, Addis Ababa institute of Technology, School of Mechanical and
Industrial Engineering, Addis Ababa, Ethiopia
ermiastes@gmail.com

Abstract. Estimation of time in project that involves several activities requires


expert’s knowledge to give an accurate estimation of project duration. In most
project, Program Evaluation and Review Technique (PERT) are used to esti-
mate completion time of project with the basic assumption of normality. This
assumption is accepted by many scholars without assessing the errors in the re-
sult. Thus, this paper examines the validity of this assumption for activities with
different probability distribution functions. The effect on the project’s comple-
tion time is addressed by considering a single probability distribution for the en-
tire activities of a network and mixed probability distribution functions within the
network, using a computer based simulation called ARENA in Villa house con-
struction project. The findings show that the project completion time for activities
that follow different probability distribution function do not follow normality.

Keywords: Project completion time (PCT), Probability distribution function


(PDF), project duration.

1 Introduction

In projects, the variability of time estimates for an activity is assumed to follow beta
distribution [1, 5] and the Project Competion Time (PCT) is assumed to follow nor-
mal distribution. Different scholars have attempt to prove the validity of this assump-
tion. Elmaghraby [4] in his work on Project Planning and Control by Network Models
recommended the use of uniform probability density function for the activity dura-
tions. Hamdy A. Taha [8] rely on the central limit theorem to postulate that the
completion time can be portrayed using a normal distribution as a function of the
cumulative mean and variance of all the a activities within the longest path. However,
Dong-Eun Lee [3] stated that a normal distribution should not always be assumed in
PCT if one wants to get a reliable result. If one assumes that PCTs are normally dis-
tributed, PERT may lead to an approximately 10 to 30% more optimistic PCT than
when activity durations are generated assuming Triangular, Uniform, Exponential and
Weibull functions. Whereas, Loostsma [9] proposed the use of the gamma probability
density function. Moreover, another assumption is that all the activities of a network
have same Probability Distribution Function (PDF). Reliable estimation of project
completion period occupies a central role in the decision making process. In the past,

© Springer International Publishing Switzerland 2015 133


A. Abraham, P. Krömer, & V. Snášel (eds.), Afro-European Conf. for Ind. Advancement,
Advances in Intelligent Systems and Computing 334, DOI: 10.1007/978-3-319-13572-4_11
134 E. Tesfaye et al.

scheduling methods by relaxing some of the restrictive assumptions of Program Eval-


uation and Review Technique (PERT) were commonly practiced. However, [3] hy-
pothesized that in practice, it is possible that different activities exhibit different
characteristics that necessitates the use of different PDFs
Several studies have been conducted on PCT of a project. Based on classification
of the literature [2], these studies can be categorized into three methods: (1) exact
methods, (2) approximation methods, and (3) simulation methods. Nowadays, simula-
tion based scheduling methods is a well-accepted technique, which gives project
planners flexibility in determining project completion time. Recently, simulation
based scheduling methods has been developed for the prediction of project comple-
tion period in use of Probability Distribution Functions (PDF) [3]. With the advent of
this method, the reliability of estimating project completion period has been
enhanced; enabling more realistic decisions to be taken.
While this technique is serving as a practical tool for scheduling project activities,
it is important to investigate its reliability as the final result is dependent on the as-
sumption that we take. Commonly the activities of a network are assumed to have a
normal distribution. Thirty years of experiences on simulation based scheduling tech-
niques relied on normality assumptions of project completion time (PCT) for the
simulation output [6, 7, 10].
It is well recognized that for a certain activity with different estimation time, eval-
uation of project completion time is commonly carried out using Program Evaluation
and Review Technique (PERT) using beta distribution. Recently, the effect of using
different probability distribution function on project completion times in simulation
based scheduling has been studied. These studies basically assume that, all activities
of a network have the same probability distribution function. However, projects with
high degree of scatter and complexity, it is possible that different activities in the
same network might have different probability distribution function.
Therefore, this paper presents a simulation of PCT that has several activities and
when the individual activities undergoes different PDF to validate the normality as-
sumption of PCT. This research attempts to address the validity of normality assump-
tion and conform the level of accuracy on the normality assumption on project
completion time.

2 Methodology

The concept of the research were based on two approaches, the first approach focused
on the theoretical simulation of a project completion time based on a hypothetical
project, with randomly generated data with different PDFs for the identical activities
of the network. Intentionally, the data were generated without altering the mean com-
pletion just simply by changing the PDFs. Keeping the mean completion time same
and differing the PDFs the results of having wide and narrow scattered activity time
were analyzed. To back the results obtained using theoretically generated data, from
A Simulation of Project Completion Probability 135

practical point of view it is important to investigate the PCT of an actual project hav-
ing wide scatter and different distribution characteristics. Since construction projects
are expected to have such characteristics, a project on a villa house was considered.
Data that are necessary for simulation were collected from nine construction compa-
nies. Every planner estimate the duration of each activity based on the company’s
trend. Finally, data were analyzed using assumed PDFs and the best fit PDF were
plotted using ARENA Vr. 14 input analyzer.

3 Simulation of Project Completion Time (PCT)

The estimation of a project time could be relatively simple as the nature of project is
of the same simple nature. However, when doing time estimations for major projects
that involve several activities, it requires a little more intellect in order to give an
accurate estimation of the time it would take to complete the project.
For projects with high degree of scatter and complexity, it is possible that different
activities in the same network might have different PDF. This theoretical simulation
analysis tries to shows the characteristics of different PDFs of individual activity time
on the total PCT.
The network shown in Figure 1 is introduced through true random number genera-
tion to demonstrate this probability distribution functions. It consists of an Activity
On-Arc (A-O-A) network with 12 activities. Each activity is assigned a most likely
duration based on random numbers generated. Two classes of random number are
generated to show both wide and narrow scatter of activity duration. To find the prob-
ability distribution function that best describes the distribution of the project competi-
tion time, analyses were made by using normal, beta, exponential and Weibull PDFs.

3.1 Simulation of Project Using Randomly Generated Data

According to the traditional PERT technique the probability of a certain project


meeting a specific schedule time can be described as follows:

Z=

Here, Z is the number of standard deviations of the due date or target date (x) lies
from the mean or expected date.
First with the normality assumption the expected completion time is determined
using generated activity network time.
136 E. Tesfaye et al.

Fig. 1. Activity Network with Narrow and Wide Scatter

The normal expected time (Te) which is equal to the sum of normal expected times
of activities on critical path. i.e. t1, t2, t3… tk are the expected times of critical path
activities, then

Te =∑ , i = 1, 2 ……, k

Thus, critical path for both narrow and wide scatter activity is the bold line as
shown in figure 1 with the expected time Te of 199.39 and 202.47 days respectively.
The project completion time with 90% confidence is 250.09 days for wide scatter
and 270.98 days for the narrow scatter activities as show in Table 1. However, the
probability of completion, assuming a normal distribution, may not hold for every
activity. Proceeding with examining the behavior of the exponential and beta distribu-
tions functions on the project completion time, on the same activity network, the ex-
pected time Te, is 173.39 and 245.33 days for wide scatter activities and 216.56 and
245.33 days for narrow scatter activities respectively.
A Simulation of Project Completion Probability 137

Table 1. Expected Time for Normal Probability Distribution Function

Probability of Project
Completion with 90 %
Expected Time (Te)
Confidence Interval
Activity

Wide Scatter Narrow Scatter


Standard Standard
Wide Narrow
Mean Deviation Mean Deviation
ac1 46.5 6.83 46.67 2.43 55.28 62.40
ac2 38.03 6.53 36.73 1.78 - -
ac3 24.23 6.08 25 2.46 - -
ac4 32.2 8.86 31.57 2.15 43.59 42.72
ac5 47.63 7.22 48.37 3.19 - -
ac6 22.6 5.98 23.07 2.56 30.28 32.20
ac7 47.1 5.04 44.63 3.03 - -
ac8 34.13 3.70 30.17 2.64 38.88 41.41
ac9 39.07 7.33 40.07 2.08 - -
ac10 33.57 6.76 32.2 2.47 42.26 43.85
ac11 33.47 4.93 35.57 2.69 39.81 48.40
ac12 30.17 5.99 32.5 2.29 - -
Expected
250.09 270.98
Time (Te)
138 E. Tesfaye et al.

ac1 acc2

ac3 acc4

ac5 aac6

ac7 acc8

ac9 ac10

ac11 ac12
Fiig. 2. Narrow Scattered Best Fit Chart

ac1 acc2

ac3 acc4

ac5 acc6

ac7 acc8

ac9 ac110

ac11 ac112

Fig. 3.. Narrow Scattered Activity Best Fit Chart


A Simulation of Project Completion Probability 139

Consequently, it is necessary to see the best fit PDFs for each activity to check the
normality assumption. The best fit PDFs is developed by using ARENA input analyz-
er to find the completion time distribution. Furthermore, the program simulates the
project completion time based on the best fit PDFs expressions.
As shown in Figure 2 and 3 the best fit PDFs for both wide and narrow scattered
activity network varied from the normality assumption where most activities show
beta PDF and there are some with Weibull, Poisson and uniform PDF which makes
the individual activity’s in the network having different PDF’s. Therefore, the next
step is to check the normality assumption of the total PCT of the network.
By simulating this best fit probability distribution expression on each activity the
project completion time was computed. The computed results for different PDF’s are
shown in Figure 4 and 5.

Fig. 4. Completion Time Distribution for narrow Scatter

Fig. 5. Completion Time Distribution for Wide Scatter


140
E. Tesfaye et al.

Fig. 6. Completion Time Distribution Simulation Network for Wide and Narrow Scatter
A Simulation of Project Completion Probability 141

3.2 Simulation of Project Using the Actual Data

The probability of the project completion time assuming the normality, beta, exponen-
tial and best fit PDF were tried to be simulated on the theoretical simulation part using
a randomly generated activity network. As it is explained above, there is a deviation
from the normality assumption where each activity follows different PDF. Therefore,
this case study tries to address ascertains that the normality assumption, which
frequently used in construction simulation studies.
This paper considers a villa house construction projects from nine construction
companies. For the purpose of simulation, only excavation and earth work, concrete
work for sub and super structure, masonry and block work activities were considered.
As shown in figure 7, the network has 21 activities and the respective mean expected
time for each activity is shown in Table 2.

Fig. 7. Network diagram for the villa house project

Table 2. Mean Expected Time for villa house project


Activity Mean Expect Activity Mean Expect Activity Mean Expect
time (Days) time (Days) time (Days)
1 3.2 8 1.6 15 6.6
2 5.1 9 1.9 16 1.7
3 5.4 10 1.6 17 2.4
4 2.8 11 2.9 18 2.7
5 3.0 12 2.7 19 3.6
6 3.5 13 4.6 20 6.6
7 4.1 14 1.0 21 8.6

Analyzing the above network with Normal, Beta, Weibull and Exponential PDF
for critical activity 1, 2, 3, 9, 13, 16, 17, 18, 20 and 21, Table 3 indicates the project
completion time for each activity and best fit PDFs is presented Figure 8. The result
indicated that each activities uses different PDF and the PCTs varies when there is
changes in PDFs. On the other hand, the normal distribution appears to best fitting for
activities only for the same PDF throughout the whole activity time.
142 E. Tesfaye et al.

Table 3. The Project Com


mpletion Time of the villa house project using different PDFs

Probability Distribution Function


Normal Beta Weibull Exponential
Project Completion
89.71 72.08 77.63 97.09
Time [days]

ac1 acc2

ac3 acc4

ac5 acc6

ac7 acc8

ac9 ac110

ac11 aac12

ac13 aac14

ac15 aac16

ac17 aac18

ac19 aac20

ac21
Fig. 8. Best Fit Probability
P Distribution Function of the 21 activities
A Simulation of Project Completion Probability 143

(a) Creatte Module (b) Process Module

Fig. 9. Snap shot of the ARENA Simulation

The procedure to modell the project network using ARENA ver. 14 simulationn is
demonstrated in figure 9. Create
C module is modified by setting constant in the T Type
field and at least 100 in thee Value field. The Max Arrivals field sets to be 10,0000 to
assure the accuracy of the histogram
h for the completion time distribution, (Williamm J.
Cosgrove, 2008)[11]. On th he process module, each activity was labeled and the acttion
set to be delay. Delay is an expression based on the best fit PDF with a unit per dayys.
144
E. Tesfaye et al.

Fig. 10. Completion Time Distribution Simulation Network for The villa house project
A Simulation of Project Completion Probability 145

As shown in the figure 9, dispose and Separate module groups are used with the
default settings because these modules have no impact on Arena network. The last
task prior to running the simulation is to construct the completion time histogram. In
the Expression field the values used to plot the histogram is based on the research
[11]. For the fields minimum, maximum, and # Cells, enter 5, 25, and 20. The first two
fields represent estimates of the range of the completion time histogram, and the last
field gives the number of time intervals on the histogram. Moreover, all the project
activities are represented in Arena by a process module as shown in the figure 10.
Consequently the results obtained from Arena Simulation which is the PCT is plot-
ted using a histogram. The simulation is repeated 10,000 times and it depicts that the
normality assumptions are invalid for different activity PDF.

4 Conclusion

In this work two methods of determining completion time distribution were demon-
strated. The first based on a hypothetically generated data and the second on an actual
construction project. The results of both simulations indicate that different activities
display different probability distribution function on a same activity network, thereby
leading to different estimation of project completion time. Thus, the current study has
demonstrated that, realistic prediction of project completion time shall be carried out
based on best fit PDFs rather than simply assuming normality.

References
1. AbouRizk, S.M., Halpin, D.W.: Statistical properties of construction duration data. J.
Constr. Eng. Manage. 111(4), 525–544 (1992)
2. Adlakha, V., Kulkarni, V.G.: A classified bibliography of research on stochastic PERT
networks. Infor. 27(3), 272–296 (1989)
3. Dong-Eun Lee, D.A.-B.: The Probability Distribution of Project Completion Times in Si-
mulation-based Scheduling. KSCE Journal of Civil Engineering 17(4), 638–645 (2013)
4. Elmaghraby, S.E.: Activity Networks: Project Planning and Control by Network Models.
Wiley, New York (1977)
5. Fente, J., Knutson, K., Schexnayder, C.: Defining a beta distribution function for construc-
tion simulation. In: Proc. 1999 Winter Simulation Conference, pp. 1010–1015. IEEE, Pis-
cataway (1999)
6. Ang, A.H.-S., Tang, W.H.: Probability concepts in engineering planning and design: Vo-
lume I - basic principles. Wiley, New York (1975)
7. Halpin, D.W., Riggs, L.S.: Planning and analysis of construction operations. Wiley, New
York (1992)
8. Taha, H.A.: Operations Research: An Introduction, 7th edn. Prentice Hall (2010)
9. Loostsma, F.A.: Network Planning with Stochastic Activity Durations: An Evaluation of
PERT. Statistica Neerlandica 20, 43–69 (1966)
10. Lu, M., AbouRizk, S.M.: Simplified CPM/PERT simulation model. Journal of Construc-
tion Engineering and Management 126(4), 219–226 (2000)
11. William, J., Cosgrove, W.: Simplifying PERT Network Simulation with ARENA. Califor-
nia Journal of Operations Management 6(1), 61–68 (2008)

You might also like